Static Multi-processor Scheduling with Ant Colony Optimisation & Local Search

Efficient multi-processor scheduling is essentially the problem of allocating a set of computational jobs to a set of processors to minimise the overall execution time. There are many variations of this problem, most of which are NP-hard, so we must rely on heuristics to solve real world problem instances. This dissertation describes several novel approaches using the ant colony optimisation (ACO) meta-heuristic and local search techniques, including tabu search, to two important versions of the problem: the static scheduling of independent jobs onto homogeneous and heterogeneous processors. Finding good schedules for jobs allocated on homogeneous processors is an important consideration if efficient use is to be made of a multiple-CPU machine, for example. An ACO algorithm to solve this problem is presented which, when combined with a fast local search procedure, can outperform traditional approaches on benchmark problems instances for the closely related bin packing problem. The algorithm cannot compete, however, with more modern specialised techniques. Scheduling jobs onto hetereogeneous processors is a more general problem which has potential applications in domains such as grid computing. A fast local search procedure for this problem is described which can quickly and effectively improve solutions built by other techniques. When used in conjunction with a well-known heuristic, Min-min, it can find shorter schedules on benchmark problems than other solution techniques found in the literature, and in significantly less time. A tabu search algorithm is also presented which can improve on solutions found by the local search procedure but takes longer. Finally a hybrid ACO algorithm which incorporates the local and tabu searches is described which outperforms both, but takes significantly longer to run.